kernel/fork.c

   1 /*
   2  *  linux/kernel/fork.c
   3  *
   4  *  Copyright (C) 1991, 1992  Linus Torvalds
   5  */
   6
   7 /*
   8  *  'fork.c' contains the help-routines for the 'fork' system call
   9  * (see also entry.S and others).
  10  * Fork is rather simple, once you get the hang of it, but the memory
  11  * management can be a bitch. See 'mm/memory.c': 'copy_page_range()'
  12  */
  13
  14 #include <linux/slab.h>
  15 #include <linux/init.h>
  16 #include <linux/unistd.h>
  17 #include <linux/module.h>
  18 #include <linux/vmalloc.h>
  19 #include <linux/completion.h>
  20 #include <linux/personality.h>
  21 #include <linux/mempolicy.h>
  22 #include <linux/sem.h>
  23 #include <linux/file.h>
  24 #include <linux/fdtable.h>
  25 #include <linux/iocontext.h>
  26 #include <linux/key.h>
  27 #include <linux/binfmts.h>
  28 #include <linux/mman.h>
  29 #include <linux/mmu_notifier.h>
  30 #include <linux/fs.h>
  31 #include <linux/nsproxy.h>
  32 #include <linux/capability.h>
  33 #include <linux/cpu.h>
  34 #include <linux/cgroup.h>
  35 #include <linux/security.h>
  36 #include <linux/hugetlb.h>
  37 #include <linux/seccomp.h>
  38 #include <linux/swap.h>
  39 #include <linux/syscalls.h>
  40 #include <linux/jiffies.h>
  41 #include <linux/futex.h>
  42 #include <linux/compat.h>
  43 #include <linux/kthread.h>
  44 #include <linux/task_io_accounting_ops.h>
  45 #include <linux/rcupdate.h>
  46 #include <linux/ptrace.h>
  47 #include <linux/mount.h>
  48 #include <linux/audit.h>
  49 #include <linux/memcontrol.h>
  50 #include <linux/ftrace.h>
  51 #include <linux/proc_fs.h>
  52 #include <linux/profile.h>
  53 #include <linux/rmap.h>
  54 #include <linux/ksm.h>
  55 #include <linux/acct.h>
  56 #include <linux/tsacct_kern.h>
  57 #include <linux/cn_proc.h>
  58 #include <linux/freezer.h>
  59 #include <linux/delayacct.h>
  60 #include <linux/taskstats_kern.h>
  61 #include <linux/random.h>
  62 #include <linux/tty.h>
  63 #include <linux/blkdev.h>
  64 #include <linux/fs_struct.h>
  65 #include <linux/magic.h>
  66 #include <linux/perf_event.h>
  67 #include <linux/posix-timers.h>
  68 #include <linux/user-return-notifier.h>
  69 #include <linux/oom.h>
  70 #include <linux/khugepaged.h>
  71 #include <linux/signalfd.h>
  72 #include <linux/uprobes.h>
  73 #include <linux/aio.h>
  74
  75 #include <asm/pgtable.h>
  76 #include <asm/pgalloc.h>
  77 #include <asm/uaccess.h>
  78 #include <asm/mmu_context.h>
  79 #include <asm/cacheflush.h>
  80 #include <asm/tlbflush.h>
  81
  82 #include <trace/events/sched.h>
  83
  84 #define CREATE_TRACE_POINTS
  85 #include <trace/events/task.h>
  86
  87 /*
  88  * Protected counters by write_lock_irq(&tasklist_lock)
  89  */
  90 unsigned long total_forks;      /* Handle normal Linux uptimes. */
  91 int nr_threads;                 /* The idle threads do not count.. */
  92
  93 int max_threads;                /* tunable limit on nr_threads */
  94
  95 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
  96
  97 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock);  /* outer */
  98
  99 #ifdef CONFIG_PROVE_RCU
 100 int lockdep_tasklist_lock_is_held(void)
 101 {
 102         return lockdep_is_held(&tasklist_lock);
 103 }
 104 EXPORT_SYMBOL_GPL(lockdep_tasklist_lock_is_held);
 105 #endif /* #ifdef CONFIG_PROVE_RCU */
 106
 107 int nr_processes(void)
 108 {
 109         int cpu;
 110         int total = 0;
 111
 112         for_each_possible_cpu(cpu)
 113                 total += per_cpu(process_counts, cpu);
 114
 115         return total;
 116 }
 117
 118 void __weak arch_release_task_struct(struct task_struct *tsk)
 119 {
 120 }
 121
 122 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
 123 static struct kmem_cache *task_struct_cachep;
 124
 125 static inline struct task_struct *alloc_task_struct_node(int node)
 126 {
 127         return kmem_cache_alloc_node(task_struct_cachep, GFP_KERNEL, node);
 128 }
 129
 130 static inline void free_task_struct(struct task_struct *tsk)
 131 {
 132         kmem_cache_free(task_struct_cachep, tsk);
 133 }
 134 #endif
 135
 136 void __weak arch_release_thread_info(struct thread_info *ti)
 137 {
 138 }
 139
 140 #ifndef CONFIG_ARCH_THREAD_INFO_ALLOCATOR
 141
 142 /*
 143  * Allocate pages if THREAD_SIZE is >= PAGE_SIZE, otherwise use a
 144  * kmemcache based allocator.
 145  */
 146 # if THREAD_SIZE >= PAGE_SIZE
 147 static struct thread_info *alloc_thread_info_node(struct task_struct *tsk,
 148                                                   int node)
 149 {
 150         struct page *page = alloc_pages_node(node, THREADINFO_GFP_ACCOUNTED,
 151                                              THREAD_SIZE_ORDER);
 152
 153         return page ? page_address(page) : NULL;
 154 }
 155
 156 static inline void free_thread_info(struct thread_info *ti)
 157 {
 158         free_memcg_kmem_pages((unsigned long)ti, THREAD_SIZE_ORDER);
 159 }
 160 # else
 161 static struct kmem_cache *thread_info_cache;
 162
 163 static struct thread_info *alloc_thread_info_node(struct task_struct *tsk,
 164                                                   int node)
 165 {
 166         return kmem_cache_alloc_node(thread_info_cache, THREADINFO_GFP, node);
 167 }
 168
 169 static void free_thread_info(struct thread_info *ti)
 170 {
 171         kmem_cache_free(thread_info_cache, ti);
 172 }
 173
 174 void thread_info_cache_init(void)
 175 {
 176         thread_info_cache = kmem_cache_create("thread_info", THREAD_SIZE,
 177                                               THREAD_SIZE, 0, NULL);
 178         BUG_ON(thread_info_cache == NULL);
 179 }
 180 # endif
 181 #endif
 182
 183 /* SLAB cache for signal_struct structures (tsk->signal) */
 184 static struct kmem_cache *signal_cachep;
 185
 186 /* SLAB cache for sighand_struct structures (tsk->sighand) */
 187 struct kmem_cache *sighand_cachep;
 188
 189 /* SLAB cache for files_struct structures (tsk->files) */
 190 struct kmem_cache *files_cachep;
 191
 192 /* SLAB cache for fs_struct structures (tsk->fs) */
 193 struct kmem_cache *fs_cachep;
 194
 195 /* SLAB cache for vm_area_struct structures */
 196 struct kmem_cache *vm_area_cachep;
 197
 198 /* SLAB cache for mm_struct structures (tsk->mm) */
 199 static struct kmem_cache *mm_cachep;
 200
 201 static void account_kernel_stack(struct thread_info *ti, int account)
 202 {
 203         struct zone *zone = page_zone(virt_to_page(ti));
 204
 205         mod_zone_page_state(zone, NR_KERNEL_STACK, account);
 206 }
 207
 208 void free_task(struct task_struct *tsk)
 209 {
 210         account_kernel_stack(tsk->stack, -1);
 211         arch_release_thread_info(tsk->stack);
 212         free_thread_info(tsk->stack);
 213         rt_mutex_debug_task_free(tsk);
 214         ftrace_graph_exit_task(tsk);
 215         put_seccomp_filter(tsk);
 216         arch_release_task_struct(tsk);
 217         free_task_struct(tsk);
 218 }
 219 EXPORT_SYMBOL(free_task);
 220
 221 static inline void free_signal_struct(struct signal_struct *sig)
 222 {
 223         taskstats_tgid_free(sig);
 224         sched_autogroup_exit(sig);
 225         kmem_cache_free(signal_cachep, sig);
 226 }
 227
 228 static inline void put_signal_struct(struct signal_struct *sig)
 229 {
 230         if (atomic_dec_and_test(&sig->sigcnt))
 231                 free_signal_struct(sig);
 232 }
 233
 234 void __put_task_struct(struct task_struct *tsk)
 235 {
 236         WARN_ON(!tsk->exit_state);
 237         WARN_ON(atomic_read(&tsk->usage));
 238         WARN_ON(tsk == current);
 239
 240         security_task_free(tsk);
 241         exit_creds(tsk);
 242         delayacct_tsk_free(tsk);
 243         put_signal_struct(tsk->signal);
 244
 245         if (!profile_handoff_task(tsk))
 246                 free_task(tsk);
 247 }
 248 EXPORT_SYMBOL_GPL(__put_task_struct);
 249
 250 void __init __weak arch_task_cache_init(void) { }
 251
 252 void __init fork_init(unsigned long mempages)
 253 {
 254 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
 255 #ifndef ARCH_MIN_TASKALIGN
 256 #define ARCH_MIN_TASKALIGN      L1_CACHE_BYTES
 257 #endif
 258         /* create a slab on which task_structs can be allocated */
 259         task_struct_cachep =
 260                 kmem_cache_create("task_struct", sizeof(struct task_struct),
 261                         ARCH_MIN_TASKALIGN, SLAB_PANIC | SLAB_NOTRACK, NULL);
 262 #endif
 263
 264         /* do the arch specific task caches init */
 265         arch_task_cache_init();
 266
 267         /*
 268          * The default maximum number of threads is set to a safe
 269          * value: the thread structures can take up at most half
 270          * of memory.
 271          */
 272         max_threads = mempages / (8 * THREAD_SIZE / PAGE_SIZE);
 273
 274         /*
 275          * we need to allow at least 20 threads to boot a system
 276          */
 277         if (max_threads < 20)
 278                 max_threads = 20;
 279
 280         init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
 281         init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
 282         init_task.signal->rlim[RLIMIT_SIGPENDING] =
 283                 init_task.signal->rlim[RLIMIT_NPROC];
 284 }
 285
 286 int __attribute__((weak)) arch_dup_task_struct(struct task_struct *dst,
 287                                                struct task_struct *src)
 288 {
 289         *dst = *src;
 290         return 0;
 291 }
 292
 293 static struct task_struct *dup_task_struct(struct task_struct *orig)
 294 {
 295         struct task_struct *tsk;
 296         struct thread_info *ti;
 297         unsigned long *stackend;
 298         int node = tsk_fork_get_node(orig);
 299         int err;
 300
 301         tsk = alloc_task_struct_node(node);
 302         if (!tsk)
 303                 return NULL;
 304
 305         ti = alloc_thread_info_node(tsk, node);
 306         if (!ti)
 307                 goto free_tsk;
 308
 309         err = arch_dup_task_struct(tsk, orig);
 310         if (err)
 311                 goto free_ti;
 312
 313         tsk->stack = ti;
 314
 315         setup_thread_stack(tsk, orig);
 316         clear_user_return_notifier(tsk);
 317         clear_tsk_need_resched(tsk);
 318         stackend = end_of_stack(tsk);
 319         *stackend = STACK_END_MAGIC;    /* for overflow detection */
 320
 321 #ifdef CONFIG_CC_STACKPROTECTOR
 322         tsk->stack_canary = get_random_int();
 323 #endif
 324
 325         /*
 326          * One for us, one for whoever does the "release_task()" (usually
 327          * parent)
 328          */
 329         atomic_set(&tsk->usage, 2);
 330 #ifdef CONFIG_BLK_DEV_IO_TRACE
 331         tsk->btrace_seq = 0;
 332 #endif
 333         tsk->splice_pipe = NULL;
 334         tsk->task_frag.page = NULL;
 335
 336         account_kernel_stack(ti, 1);
 337
 338         return tsk;
 339
 340 free_ti:
 341         free_thread_info(ti);
 342 free_tsk:
 343         free_task_struct(tsk);
 344         return NULL;
 345 }
 346
 347 #ifdef CONFIG_MMU
 348 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
 349 {
 350         struct vm_area_struct *mpnt, *tmp, *prev, **pprev;
 351         struct rb_node **rb_link, *rb_parent;
 352         int retval;
 353         unsigned long charge;
 354
 355         uprobe_start_dup_mmap();
 356         down_write(&oldmm->mmap_sem);
 357         flush_cache_dup_mm(oldmm);
 358         uprobe_dup_mmap(oldmm, mm);
 359         /*
 360          * Not linked in yet - no deadlock potential:
 361          */
 362         down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING);
 363
 364         mm->locked_vm = 0;
 365         mm->mmap = NULL;
 366         mm->mmap_cache = NULL;
 367         mm->map_count = 0;
 368         cpumask_clear(mm_cpumask(mm));
 369         mm->mm_rb = RB_ROOT;
 370         rb_link = &mm->mm_rb.rb_node;
 371         rb_parent = NULL;
 372         pprev = &mm->mmap;
 373         retval = ksm_fork(mm, oldmm);
 374         if (retval)
 375                 goto out;
 376         retval = khugepaged_fork(mm, oldmm);
 377         if (retval)
 378                 goto out;
 379
 380         prev = NULL;
 381         for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
 382                 struct file *file;
 383
 384                 if (mpnt->vm_flags & VM_DONTCOPY) {
 385                         vm_stat_account(mm, mpnt->vm_flags, mpnt->vm_file,
 386                                                         -vma_pages(mpnt));
 387                         continue;
 388                 }
 389                 charge = 0;
 390                 if (mpnt->vm_flags & VM_ACCOUNT) {
 391                         unsigned long len = vma_pages(mpnt);
 392
 393                         if (security_vm_enough_memory_mm(oldmm, len)) /* sic */
 394                                 goto fail_nomem;
 395                         charge = len;
 396                 }
 397                 tmp = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
 398                 if (!tmp)
 399                         goto fail_nomem;
 400                 *tmp = *mpnt;
 401                 INIT_LIST_HEAD(&tmp->anon_vma_chain);
 402                 retval = vma_dup_policy(mpnt, tmp);
 403                 if (retval)
 404                         goto fail_nomem_policy;
 405                 tmp->vm_mm = mm;
 406                 if (anon_vma_fork(tmp, mpnt))
 407                         goto fail_nomem_anon_vma_fork;
 408                 tmp->vm_flags &= ~VM_LOCKED;
 409                 tmp->vm_next = tmp->vm_prev = NULL;
 410                 file = tmp->vm_file;
 411                 if (file) {
 412                         struct inode *inode = file_inode(file);
 413                         struct address_space *mapping = file->f_mapping;
 414
 415                         get_file(file);
 416                         if (tmp->vm_flags & VM_DENYWRITE)
 417                                 atomic_dec(&inode->i_writecount);
 418                         mutex_lock(&mapping->i_mmap_mutex);
 419                         if (tmp->vm_flags & VM_SHARED)
 420                                 mapping->i_mmap_writable++;
 421                         flush_dcache_mmap_lock(mapping);
 422                         /* insert tmp into the share list, just after mpnt */
 423                         if (unlikely(tmp->vm_flags & VM_NONLINEAR))
 424                                 vma_nonlinear_insert(tmp,
 425                                                 &mapping->i_mmap_nonlinear);
 426                         else
 427                                 vma_interval_tree_insert_after(tmp, mpnt,
 428                                                         &mapping->i_mmap);
 429                         flush_dcache_mmap_unlock(mapping);
 430                         mutex_unlock(&mapping->i_mmap_mutex);
 431                 }
 432
 433                 /*
 434                  * Clear hugetlb-related page reserves for children. This only
 435                  * affects MAP_PRIVATE mappings. Faults generated by the child
 436                  * are not guaranteed to succeed, even if read-only
 437                  */
 438                 if (is_vm_hugetlb_page(tmp))
 439                         reset_vma_resv_huge_pages(tmp);
 440
 441                 /*
 442                  * Link in the new vma and copy the page table entries.
 443                  */
 444                 *pprev = tmp;
 445                 pprev = &tmp->vm_next;
 446                 tmp->vm_prev = prev;
 447                 prev = tmp;
 448
 449                 __vma_link_rb(mm, tmp, rb_link, rb_parent);
 450                 rb_link = &tmp->vm_rb.rb_right;
 451                 rb_parent = &tmp->vm_rb;
 452
 453                 mm->map_count++;
 454                 retval = copy_page_range(mm, oldmm, mpnt);
 455
 456                 if (tmp->vm_ops && tmp->vm_ops->open)
 457                         tmp->vm_ops->open(tmp);
 458
 459                 if (retval)
 460                         goto out;
 461         }
 462         /* a new mm has just been created */
 463         arch_dup_mmap(oldmm, mm);
 464         retval = 0;
 465 out:
 466         up_write(&mm->mmap_sem);
 467         flush_tlb_mm(oldmm);
 468         up_write(&oldmm->mmap_sem);
 469         uprobe_end_dup_mmap();
 470         return retval;
 471 fail_nomem_anon_vma_fork:
 472         mpol_put(vma_policy(tmp));
 473 fail_nomem_policy:
 474         kmem_cache_free(vm_area_cachep, tmp);
 475 fail_nomem:
 476         retval = -ENOMEM;
 477         vm_unacct_memory(charge);
 478         goto out;
 479 }
 480
 481 static inline int mm_alloc_pgd(struct mm_struct *mm)
 482 {
 483         mm->pgd = pgd_alloc(mm);
 484         if (unlikely(!mm->pgd))
 485                 return -ENOMEM;
 486         return 0;
 487 }
 488
 489 static inline void mm_free_pgd(struct mm_struct *mm)
 490 {
 491         pgd_free(mm, mm->pgd);
 492 }
 493 #else
 494 #define dup_mmap(mm, oldmm)     (0)
 495 #define mm_alloc_pgd(mm)        (0)
 496 #define mm_free_pgd(mm)
 497 #endif /* CONFIG_MMU */
 498
 499 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
 500
 501 #define allocate_mm()   (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
 502 #define free_mm(mm)     (kmem_cache_free(mm_cachep, (mm)))
 503
 504 static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
 505
 506 static int __init coredump_filter_setup(char *s)
 507 {
 508         default_dump_filter =
 509                 (simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
 510                 MMF_DUMP_FILTER_MASK;
 511         return 1;
 512 }
 513
 514 __setup("coredump_filter=", coredump_filter_setup);
 515
 516 #include <linux/init_task.h>
 517
 518 static void mm_init_aio(struct mm_struct *mm)
 519 {
 520 #ifdef CONFIG_AIO
 521         spin_lock_init(&mm->ioctx_lock);
 522         mm->ioctx_table = NULL;
 523 #endif
 524 }
 525
 526 static struct mm_struct *mm_init(struct mm_struct *mm, struct task_struct *p)
 527 {
 528         atomic_set(&mm->mm_users, 1);
 529         atomic_set(&mm->mm_count, 1);
 530         init_rwsem(&mm->mmap_sem);
 531         INIT_LIST_HEAD(&mm->mmlist);
 532         mm->flags = (current->mm) ?
 533                 (current->mm->flags & MMF_INIT_MASK) : default_dump_filter;
 534         mm->core_state = NULL;
 535         atomic_long_set(&mm->nr_ptes, 0);
 536         memset(&mm->rss_stat, 0, sizeof(mm->rss_stat));
 537         spin_lock_init(&mm->page_table_lock);
 538         mm_init_aio(mm);
 539         mm_init_owner(mm, p);
 540         clear_tlb_flush_pending(mm);
 541
 542         if (likely(!mm_alloc_pgd(mm))) {
 543                 mm->def_flags = 0;
 544                 mmu_notifier_mm_init(mm);
 545                 return mm;
 546         }
 547
 548         free_mm(mm);
 549         return NULL;
 550 }
 551
 552 static void check_mm(struct mm_struct *mm)
 553 {
 554         int i;
 555
 556         for (i = 0; i < NR_MM_COUNTERS; i++) {
 557                 long x = atomic_long_read(&mm->rss_stat.count[i]);
 558
 559                 if (unlikely(x))
 560                         printk(KERN_ALERT "BUG: Bad rss-counter state "
 561                                           "mm:%p idx:%d val:%ld\n", mm, i, x);
 562         }
 563
 564 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
 565         VM_BUG_ON(mm->pmd_huge_pte);
 566 #endif
 567 }
 568
 569 /*
 570  * Allocate and initialize an mm_struct.
 571  */
 572 struct mm_struct *mm_alloc(void)
 573 {
 574         struct mm_struct *mm;
 575
 576         mm = allocate_mm();
 577         if (!mm)
 578                 return NULL;
 579
 580         memset(mm, 0, sizeof(*mm));
 581         mm_init_cpumask(mm);
 582         return mm_init(mm, current);
 583 }
 584
 585 /*
 586  * Called when the last reference to the mm
 587  * is dropped: either by a lazy thread or by
 588  * mmput. Free the page directory and the mm.
 589  */
 590 void __mmdrop(struct mm_struct *mm)
 591 {
 592         BUG_ON(mm == &init_mm);
 593         mm_free_pgd(mm);
 594         destroy_context(mm);
 595         mmu_notifier_mm_destroy(mm);
 596         check_mm(mm);
 597         free_mm(mm);
 598 }
 599 EXPORT_SYMBOL_GPL(__mmdrop);
 600
 601 /*
 602  * Decrement the use count and release all resources for an mm.
 603  */
 604 void mmput(struct mm_struct *mm)
 605 {
 606         might_sleep();
 607
 608         if (atomic_dec_and_test(&mm->mm_users)) {
 609                 uprobe_clear_state(mm);
 610                 exit_aio(mm);
 611                 ksm_exit(mm);
 612                 khugepaged_exit(mm); /* must run before exit_mmap */
 613                 exit_mmap(mm);
 614                 set_mm_exe_file(mm, NULL);
 615                 if (!list_empty(&mm->mmlist)) {
 616                         spin_lock(&mmlist_lock);
 617                         list_del(&mm->mmlist);
 618                         spin_unlock(&mmlist_lock);
 619                 }
 620                 if (mm->binfmt)
 621                         module_put(mm->binfmt->module);
 622                 mmdrop(mm);
 623         }
 624 }
 625 EXPORT_SYMBOL_GPL(mmput);
 626
 627 void set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
 628 {
 629         if (new_exe_file)
 630                 get_file(new_exe_file);
 631         if (mm->exe_file)
 632                 fput(mm->exe_file);
 633         mm->exe_file = new_exe_file;
 634 }
 635
 636 struct file *get_mm_exe_file(struct mm_struct *mm)
 637 {
 638         struct file *exe_file;
 639
 640         /* We need mmap_sem to protect against races with removal of exe_file */
 641         down_read(&mm->mmap_sem);
 642         exe_file = mm->exe_file;
 643         if (exe_file)
 644                 get_file(exe_file);
 645         up_read(&mm->mmap_sem);
 646         return exe_file;
 647 }
 648
 649 static void dup_mm_exe_file(struct mm_struct *oldmm, struct mm_struct *newmm)
 650 {
 651         /* It's safe to write the exe_file pointer without exe_file_lock because
 652          * this is called during fork when the task is not yet in /proc */
 653         newmm->exe_file = get_mm_exe_file(oldmm);
 654 }
 655
 656 /**
 657  * get_task_mm - acquire a reference to the task's mm
 658  *
 659  * Returns %NULL if the task has no mm.  Checks PF_KTHREAD (meaning
 660  * this kernel workthread has transiently adopted a user mm with use_mm,
 661  * to do its AIO) is not set and if so returns a reference to it, after
 662  * bumping up the use count.  User must release the mm via mmput()
 663  * after use.  Typically used by /proc and ptrace.
 664  */
 665 struct mm_struct *get_task_mm(struct task_struct *task)
 666 {
 667         struct mm_struct *mm;
 668
 669         task_lock(task);
 670         mm = task->mm;
 671         if (mm) {
 672                 if (task->flags & PF_KTHREAD)
 673                         mm = NULL;
 674                 else
 675                         atomic_inc(&mm->mm_users);
 676         }
 677         task_unlock(task);
 678         return mm;
 679 }
 680 EXPORT_SYMBOL_GPL(get_task_mm);
 681
 682 struct mm_struct *mm_access(struct task_struct *task, unsigned int mode)
 683 {
 684         struct mm_struct *mm;
 685         int err;
 686
 687         err =  mutex_lock_killable(&task->signal->cred_guard_mutex);
 688         if (err)
 689                 return ERR_PTR(err);
 690
 691         mm = get_task_mm(task);
 692         if (mm && mm != current->mm &&
 693                         !ptrace_may_access(task, mode)) {
 694                 mmput(mm);
 695                 mm = ERR_PTR(-EACCES);
 696         }
 697         mutex_unlock(&task->signal->cred_guard_mutex);
 698
 699         return mm;
 700 }
 701
 702 static void complete_vfork_done(struct task_struct *tsk)
 703 {
 704         struct completion *vfork;
 705
 706         task_lock(tsk);
 707         vfork = tsk->vfork_done;
 708         if (likely(vfork)) {
 709                 tsk->vfork_done = NULL;
 710                 complete(vfork);
 711         }
 712         task_unlock(tsk);
 713 }
 714
 715 static int wait_for_vfork_done(struct task_struct *child,
 716                                 struct completion *vfork)
 717 {
 718         int killed;
 719
 720         freezer_do_not_count();
 721         killed = wait_for_completion_killable(vfork);
 722         freezer_count();
 723
 724         if (killed) {
 725                 task_lock(child);
 726                 child->vfork_done = NULL;
 727                 task_unlock(child);
 728         }
 729
 730         put_task_struct(child);
 731         return killed;
 732 }
 733
 734 /* Please note the differences between mmput and mm_release.
 735  * mmput is called whenever we stop holding onto a mm_struct,
 736  * error success whatever.
 737  *
 738  * mm_release is called after a mm_struct has been removed
 739  * from the current process.
 740  *
 741  * This difference is important for error handling, when we
 742  * only half set up a mm_struct for a new process and need to restore
 743  * the old one.  Because we mmput the new mm_struct before
 744  * restoring the old one. . .
 745  * Eric Biederman 10 January 1998
 746  */
 747 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
 748 {
 749         /* Get rid of any futexes when releasing the mm */
 750 #ifdef CONFIG_FUTEX
 751         if (unlikely(tsk->robust_list)) {
 752                 exit_robust_list(tsk);
 753                 tsk->robust_list = NULL;
 754         }
 755 #ifdef CONFIG_COMPAT
 756         if (unlikely(tsk->compat_robust_list)) {
 757                 compat_exit_robust_list(tsk);
 758                 tsk->compat_robust_list = NULL;
 759         }
 760 #endif
 761         if (unlikely(!list_empty(&tsk->pi_state_list)))
 762                 exit_pi_state_list(tsk);
 763 #endif
 764
 765         uprobe_free_utask(tsk);
 766
 767         /* Get rid of any cached register state */
 768         deactivate_mm(tsk, mm);
 769
 770         /*
 771          * If we're exiting normally, clear a user-space tid field if
 772          * requested.  We leave this alone when dying by signal, to leave
 773          * the value intact in a core dump, and to save the unnecessary
 774          * trouble, say, a killed vfork parent shouldn't touch this mm.
 775          * Userland only wants this done for a sys_exit.
 776          */
 777         if (tsk->clear_child_tid) {
 778                 if (!(tsk->flags & PF_SIGNALED) &&
 779                     atomic_read(&mm->mm_users) > 1) {
 780                         /*
 781                          * We don't check the error code - if userspace has
 782                          * not set up a proper pointer then tough luck.
 783                          */
 784                         put_user(0, tsk->clear_child_tid);
 785                         sys_futex(tsk->clear_child_tid, FUTEX_WAKE,
 786                                         1, NULL, NULL, 0);
 787                 }
 788                 tsk->clear_child_tid = NULL;
 789         }
 790
 791         /*
 792          * All done, finally we can wake up parent and return this mm to him.
 793          * Also kthread_stop() uses this completion for synchronization.
 794          */
 795         if (tsk->vfork_done)
 796                 complete_vfork_done(tsk);
 797 }
 798
 799 /*
 800  * Allocate a new mm structure and copy contents from the
 801  * mm structure of the passed in task structure.
 802  */
 803 static struct mm_struct *dup_mm(struct task_struct *tsk)
 804 {
 805         struct mm_struct *mm, *oldmm = current->mm;
 806         int err;
 807
 808         mm = allocate_mm();
 809         if (!mm)
 810                 goto fail_nomem;
 811
 812         memcpy(mm, oldmm, sizeof(*mm));
 813         mm_init_cpumask(mm);
 814
 815 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
 816         mm->pmd_huge_pte = NULL;
 817 #endif
 818         if (!mm_init(mm, tsk))
 819                 goto fail_nomem;
 820
 821         if (init_new_context(tsk, mm))
 822                 goto fail_nocontext;
 823
 824         dup_mm_exe_file(oldmm, mm);
 825
 826         err = dup_mmap(mm, oldmm);
 827         if (err)
 828                 goto free_pt;
 829
 830         mm->hiwater_rss = get_mm_rss(mm);
 831         mm->hiwater_vm = mm->total_vm;
 832
 833         if (mm->binfmt && !try_module_get(mm->binfmt->module))
 834                 goto free_pt;
 835
 836         return mm;
 837
 838 free_pt:
 839         /* don't put binfmt in mmput, we haven't got module yet */
 840         mm->binfmt = NULL;
 841         mmput(mm);
 842
 843 fail_nomem:
 844         return NULL;
 845
 846 fail_nocontext:
 847         /*
 848          * If init_new_context() failed, we cannot use mmput() to free the mm
 849          * because it calls destroy_context()
 850          */
 851         mm_free_pgd(mm);
 852         free_mm(mm);
 853         return NULL;
 854 }
 855
 856 static int copy_mm(unsigned long clone_flags, struct task_struct *tsk)
 857 {
 858         struct mm_struct *mm, *oldmm;
 859         int retval;
 860
 861         tsk->min_flt = tsk->maj_flt = 0;
 862         tsk->nvcsw = tsk->nivcsw = 0;
 863 #ifdef CONFIG_DETECT_HUNG_TASK
 864         tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
 865 #endif
 866
 867         tsk->mm = NULL;
 868         tsk->active_mm = NULL;
 869
 870         /*
 871          * Are we cloning a kernel thread?
 872          *
 873          * We need to steal a active VM for that..
 874          */
 875         oldmm = current->mm;
 876         if (!oldmm)
 877                 return 0;
 878
 879         if (clone_flags & CLONE_VM) {
 880                 atomic_inc(&oldmm->mm_users);
 881                 mm = oldmm;
 882                 goto good_mm;
 883         }
 884
 885         retval = -ENOMEM;
 886         mm = dup_mm(tsk);
 887         if (!mm)
 888                 goto fail_nomem;
 889
 890 good_mm:
 891         tsk->mm = mm;
 892         tsk->active_mm = mm;
 893         return 0;
 894
 895 fail_nomem:
 896         return retval;
 897 }
 898
 899 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
 900 {
 901         struct fs_struct *fs = current->fs;
 902         if (clone_flags & CLONE_FS) {
 903                 /* tsk->fs is already what we want */
 904                 spin_lock(&fs->lock);
 905                 if (fs->in_exec) {
 906                         spin_unlock(&fs->lock);
 907                         return -EAGAIN;
 908                 }
 909                 fs->users++;
 910                 spin_unlock(&fs->lock);
 911                 return 0;
 912         }
 913         tsk->fs = copy_fs_struct(fs);
 914         if (!tsk->fs)
 915                 return -ENOMEM;
 916         return 0;
 917 }
 918
 919 static int copy_files(unsigned long clone_flags, struct task_struct *tsk)
 920 {
 921         struct files_struct *oldf, *newf;
 922         int error = 0;
 923
 924         /*
 925          * A background process may not have any files ...
 926          */
 927         oldf = current->files;
 928         if (!oldf)
 929                 goto out;
 930
 931         if (clone_flags & CLONE_FILES) {
 932                 atomic_inc(&oldf->count);
 933                 goto out;
 934         }
 935
 936         newf = dup_fd(oldf, &error);
 937         if (!newf)
 938                 goto out;
 939
 940         tsk->files = newf;
 941         error = 0;
 942 out:
 943         return error;
 944 }
 945
 946 static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
 947 {
 948 #ifdef CONFIG_BLOCK
 949         struct io_context *ioc = current->io_context;
 950         struct io_context *new_ioc;
 951
 952         if (!ioc)
 953                 return 0;
 954         /*
 955          * Share io context with parent, if CLONE_IO is set
 956          */
 957         if (clone_flags & CLONE_IO) {
 958                 ioc_task_link(ioc);
 959                 tsk->io_context = ioc;
 960         } else if (ioprio_valid(ioc->ioprio)) {
 961                 new_ioc = get_task_io_context(tsk, GFP_KERNEL, NUMA_NO_NODE);
 962                 if (unlikely(!new_ioc))
 963                         return -ENOMEM;
 964
 965                 new_ioc->ioprio = ioc->ioprio;
 966                 put_io_context(new_ioc);
 967         }
 968 #endif
 969         return 0;
 970 }
 971
 972 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
 973 {
 974         struct sighand_struct *sig;
 975
 976         if (clone_flags & CLONE_SIGHAND) {
 977                 atomic_inc(&current->sighand->count);
 978                 return 0;
 979         }
 980         sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
 981         rcu_assign_pointer(tsk->sighand, sig);
 982         if (!sig)
 983                 return -ENOMEM;
 984         atomic_set(&sig->count, 1);
 985         memcpy(sig->action, current->sighand->action, sizeof(sig->action));
 986         return 0;
 987 }
 988
 989 void __cleanup_sighand(struct sighand_struct *sighand)
 990 {
 991         if (atomic_dec_and_test(&sighand->count)) {
 992                 signalfd_cleanup(sighand);
 993                 kmem_cache_free(sighand_cachep, sighand);
 994         }
 995 }
 996
 997
 998 /*
 999  * Initialize POSIX timer handling for a thread group.
1000  */
1001 static void posix_cpu_timers_init_group(struct signal_struct *sig)
1002 {
1003         unsigned long cpu_limit;
1004
1005         /* Thread group counters. */
1006         thread_group_cputime_init(sig);
1007
1008         cpu_limit = ACCESS_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
1009         if (cpu_limit != RLIM_INFINITY) {
1010                 sig->cputime_expires.prof_exp = secs_to_cputime(cpu_limit);
1011                 sig->cputimer.running = 1;
1012         }
1013
1014         /* The timer lists. */
1015         INIT_LIST_HEAD(&sig->cpu_timers[0]);
1016         INIT_LIST_HEAD(&sig->cpu_timers[1]);
1017         INIT_LIST_HEAD(&sig->cpu_timers[2]);
1018 }
1019
1020 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
1021 {
1022         struct signal_struct *sig;
1023
1024         if (clone_flags & CLONE_THREAD)
1025                 return 0;
1026
1027         sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
1028         tsk->signal = sig;
1029         if (!sig)
1030                 return -ENOMEM;
1031
1032         sig->nr_threads = 1;
1033         atomic_set(&sig->live, 1);
1034         atomic_set(&sig->sigcnt, 1);
1035
1036         /* list_add(thread_node, thread_head) without INIT_LIST_HEAD() */
1037         sig->thread_head = (struct list_head)LIST_HEAD_INIT(tsk->thread_node);
1038         tsk->thread_node = (struct list_head)LIST_HEAD_INIT(sig->thread_head);
1039
1040         init_waitqueue_head(&sig->wait_chldexit);
1041         sig->curr_target = tsk;
1042         init_sigpending(&sig->shared_pending);
1043         INIT_LIST_HEAD(&sig->posix_timers);
1044
1045         hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1046         sig->real_timer.function = it_real_fn;
1047
1048         task_lock(current->group_leader);
1049         memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
1050         task_unlock(current->group_leader);
1051
1052         posix_cpu_timers_init_group(sig);
1053
1054         tty_audit_fork(sig);
1055         sched_autogroup_fork(sig);
1056
1057 #ifdef CONFIG_CGROUPS
1058         init_rwsem(&sig->group_rwsem);
1059 #endif
1060
1061         sig->oom_score_adj = current->signal->oom_score_adj;
1062         sig->oom_score_adj_min = current->signal->oom_score_adj_min;
1063
1064         sig->has_child_subreaper = current->signal->has_child_subreaper ||
1065                                    current->signal->is_child_subreaper;
1066
1067         mutex_init(&sig->cred_guard_mutex);
1068
1069         return 0;
1070 }
1071
1072 static void copy_flags(unsigned long clone_flags, struct task_struct *p)
1073 {
1074         unsigned long new_flags = p->flags;
1075
1076         new_flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER);
1077         new_flags |= PF_FORKNOEXEC;
1078         p->flags = new_flags;
1079 }
1080
1081 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
1082 {
1083         current->clear_child_tid = tidptr;
1084
1085         return task_pid_vnr(current);
1086 }
1087
1088 static void rt_mutex_init_task(struct task_struct *p)
1089 {
1090         raw_spin_lock_init(&p->pi_lock);
1091 #ifdef CONFIG_RT_MUTEXES
1092         p->pi_waiters = RB_ROOT;
1093         p->pi_waiters_leftmost = NULL;
1094         p->pi_blocked_on = NULL;
1095         p->pi_top_task = NULL;
1096 #endif
1097 }
1098
1099 #ifdef CONFIG_MM_OWNER
1100 void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
1101 {
1102         mm->owner = p;
1103 }
1104 #endif /* CONFIG_MM_OWNER */
1105
1106 /*
1107  * Initialize POSIX timer handling for a single task.
1108  */
1109 static void posix_cpu_timers_init(struct task_struct *tsk)
1110 {
1111         tsk->cputime_expires.prof_exp = 0;
1112         tsk->cputime_expires.virt_exp = 0;
1113         tsk->cputime_expires.sched_exp = 0;
1114         INIT_LIST_HEAD(&tsk->cpu_timers[0]);
1115         INIT_LIST_HEAD(&tsk->cpu_timers[1]);
1116         INIT_LIST_HEAD(&tsk->cpu_timers[2]);
1117 }
1118
1119 static inline void
1120 init_task_pid(struct task_struct *task, enum pid_type type, struct pid *pid)
1121 {
1122          task->pids[type].pid = pid;
1123 }
1124
1125 /*
1126  * This creates a new process as a copy of the old one,
1127  * but does not actually start it yet.
1128  *
1129  * It copies the registers, and all the appropriate
1130  * parts of the process environment (as per the clone
1131  * flags). The actual kick-off is left to the caller.
1132  */
1133 static struct task_struct *copy_process(unsigned long clone_flags,
1134                                         unsigned long stack_start,
1135                                         unsigned long stack_size,
1136                                         int __user *child_tidptr,
1137                                         struct pid *pid,
1138                                         int trace)
1139 {
1140         int retval;
1141         struct task_struct *p;
1142
1143         if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
1144                 return ERR_PTR(-EINVAL);
1145
1146         if ((clone_flags & (CLONE_NEWUSER|CLONE_FS)) == (CLONE_NEWUSER|CLONE_FS))
1147                 return ERR_PTR(-EINVAL);
1148
1149         /*
1150          * Thread groups must share signals as well, and detached threads
1151          * can only be started up within the thread group.
1152          */
1153         if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
1154                 return ERR_PTR(-EINVAL);
1155
1156         /*
1157          * Shared signal handlers imply shared VM. By way of the above,
1158          * thread groups also imply shared VM. Blocking this case allows
1159          * for various simplifications in other code.
1160          */
1161         if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
1162                 return ERR_PTR(-EINVAL);
1163
1164         /*
1165          * Siblings of global init remain as zombies on exit since they are
1166          * not reaped by their parent (swapper). To solve this and to avoid
1167          * multi-rooted process trees, prevent global and container-inits
1168          * from creating siblings.
1169          */
1170         if ((clone_flags & CLONE_PARENT) &&
1171                                 current->signal->flags & SIGNAL_UNKILLABLE)
1172                 return ERR_PTR(-EINVAL);
1173
1174         /*
1175          * If the new process will be in a different pid or user namespace
1176          * do not allow it to share a thread group or signal handlers or
1177          * parent with the forking task.
1178          */
1179         if (clone_flags & CLONE_SIGHAND) {
1180                 if ((clone_flags & (CLONE_NEWUSER | CLONE_NEWPID)) ||
1181                     (task_active_pid_ns(current) !=
1182                                 current->nsproxy->pid_ns_for_children))
1183                         return ERR_PTR(-EINVAL);
1184         }
1185
1186         retval = security_task_create(clone_flags);
1187         if (retval)
1188                 goto fork_out;
1189
1190         retval = -ENOMEM;
1191         p = dup_task_struct(current);
1192         if (!p)
1193                 goto fork_out;
1194
1195         ftrace_graph_init_task(p);
1196         get_seccomp_filter(p);
1197
1198         rt_mutex_init_task(p);
1199
1200 #ifdef CONFIG_PROVE_LOCKING
1201         DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
1202         DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
1203 #endif
1204         retval = -EAGAIN;
1205         if (atomic_read(&p->real_cred->user->processes) >=
1206                         task_rlimit(p, RLIMIT_NPROC)) {
1207                 if (p->real_cred->user != INIT_USER &&
1208                     !capable(CAP_SYS_RESOURCE) && !capable(CAP_SYS_ADMIN))
1209                         goto bad_fork_free;
1210         }
1211         current->flags &= ~PF_NPROC_EXCEEDED;
1212
1213         retval = copy_creds(p, clone_flags);
1214         if (retval < 0)
1215                 goto bad_fork_free;
1216
1217         /*
1218          * If multiple threads are within copy_process(), then this check
1219          * triggers too late. This doesn't hurt, the check is only there
1220          * to stop root fork bombs.
1221          */
1222         retval = -EAGAIN;
1223         if (nr_threads >= max_threads)
1224                 goto bad_fork_cleanup_count;
1225
1226         if (!try_module_get(task_thread_info(p)->exec_domain->module))
1227                 goto bad_fork_cleanup_count;
1228
1229         delayacct_tsk_init(p);  /* Must remain after dup_task_struct() */
1230         copy_flags(clone_flags, p);
1231         INIT_LIST_HEAD(&p->children);
1232         INIT_LIST_HEAD(&p->sibling);
1233         rcu_copy_process(p);
1234         p->vfork_done = NULL;
1235         spin_lock_init(&p->alloc_lock);
1236
1237         init_sigpending(&p->pending);
1238
1239         p->utime = p->stime = p->gtime = 0;
1240         p->utimescaled = p->stimescaled = 0;
1241 #ifndef CONFIG_VIRT_CPU_ACCOUNTING_NATIVE
1242         p->prev_cputime.utime = p->prev_cputime.stime = 0;
1243 #endif
1244 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
1245         seqlock_init(&p->vtime_seqlock);
1246         p->vtime_snap = 0;
1247         p->vtime_snap_whence = VTIME_SLEEPING;
1248 #endif
1249
1250 #if defined(SPLIT_RSS_COUNTING)
1251         memset(&p->rss_stat, 0, sizeof(p->rss_stat));
1252 #endif
1253
1254         p->default_timer_slack_ns = current->timer_slack_ns;
1255
1256         task_io_accounting_init(&p->ioac);
1257         acct_clear_integrals(p);
1258
1259         posix_cpu_timers_init(p);
1260
1261         do_posix_clock_monotonic_gettime(&p->start_time);
1262         p->real_start_time = p->start_time;
1263         monotonic_to_bootbased(&p->real_start_time);
1264         p->io_context = NULL;
1265         p->audit_context = NULL;
1266         if (clone_flags & CLONE_THREAD)
1267                 threadgroup_change_begin(current);
1268         cgroup_fork(p);
1269 #ifdef CONFIG_NUMA
1270         p->mempolicy = mpol_dup(p->mempolicy);
1271         if (IS_ERR(p->mempolicy)) {
1272                 retval = PTR_ERR(p->mempolicy);
1273                 p->mempolicy = NULL;
1274                 goto bad_fork_cleanup_cgroup;
1275         }
1276         mpol_fix_fork_child_flag(p);
1277 #endif
1278 #ifdef CONFIG_CPUSETS
1279         p->cpuset_mem_spread_rotor = NUMA_NO_NODE;
1280         p->cpuset_slab_spread_rotor = NUMA_NO_NODE;
1281         seqcount_init(&p->mems_allowed_seq);
1282 #endif
1283 #ifdef CONFIG_TRACE_IRQFLAGS
1284         p->irq_events = 0;
1285         p->hardirqs_enabled = 0;
1286         p->hardirq_enable_ip = 0;
1287         p->hardirq_enable_event = 0;
1288         p->hardirq_disable_ip = _THIS_IP_;
1289         p->hardirq_disable_event = 0;
1290         p->softirqs_enabled = 1;
1291         p->softirq_enable_ip = _THIS_IP_;
1292         p->softirq_enable_event = 0;
1293         p->softirq_disable_ip = 0;
1294         p->softirq_disable_event = 0;
1295         p->hardirq_context = 0;
1296         p->softirq_context = 0;
1297 #endif
1298 #ifdef CONFIG_LOCKDEP
1299         p->lockdep_depth = 0; /* no locks held yet */
1300         p->curr_chain_key = 0;
1301         p->lockdep_recursion = 0;
1302 #endif
1303
1304 #ifdef CONFIG_DEBUG_MUTEXES
1305         p->blocked_on = NULL; /* not blocked yet */
1306 #endif
1307 #ifdef CONFIG_MEMCG
1308         p->memcg_batch.do_batch = 0;
1309         p->memcg_batch.memcg = NULL;
1310 #endif
1311 #ifdef CONFIG_BCACHE
1312         p->sequential_io        = 0;
1313         p->sequential_io_avg    = 0;
1314 #endif
1315
1316         /* Perform scheduler related setup. Assign this task to a CPU. */
1317         retval = sched_fork(clone_flags, p);
1318         if (retval)
1319                 goto bad_fork_cleanup_policy;
1320
1321         retval = perf_event_init_task(p);
1322         if (retval)
1323                 goto bad_fork_cleanup_policy;
1324         retval = audit_alloc(p);
1325         if (retval)
1326                 goto bad_fork_cleanup_policy;
1327         /* copy all the process information */
1328         retval = copy_semundo(clone_flags, p);
1329         if (retval)
1330                 goto bad_fork_cleanup_audit;
1331         retval = copy_files(clone_flags, p);
1332         if (retval)
1333                 goto bad_fork_cleanup_semundo;
1334         retval = copy_fs(clone_flags, p);
1335         if (retval)
1336                 goto bad_fork_cleanup_files;
1337         retval = copy_sighand(clone_flags, p);
1338         if (retval)
1339                 goto bad_fork_cleanup_fs;
1340         retval = copy_signal(clone_flags, p);
1341         if (retval)
1342                 goto bad_fork_cleanup_sighand;
1343         retval = copy_mm(clone_flags, p);
1344         if (retval)
1345                 goto bad_fork_cleanup_signal;
1346         retval = copy_namespaces(clone_flags, p);
1347         if (retval)
1348                 goto bad_fork_cleanup_mm;
1349         retval = copy_io(clone_flags, p);
1350         if (retval)
1351                 goto bad_fork_cleanup_namespaces;
1352         retval = copy_thread(clone_flags, stack_start, stack_size, p);
1353         if (retval)
1354                 goto bad_fork_cleanup_io;
1355
1356         if (pid != &init_struct_pid) {
1357                 retval = -ENOMEM;
1358                 pid = alloc_pid(p->nsproxy->pid_ns_for_children);
1359                 if (!pid)
1360                         goto bad_fork_cleanup_io;
1361         }
1362
1363         p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1364         /*
1365          * Clear TID on mm_release()?
1366          */
1367         p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr : NULL;
1368 #ifdef CONFIG_BLOCK
1369         p->plug = NULL;
1370 #endif
1371 #ifdef CONFIG_FUTEX
1372         p->robust_list = NULL;
1373 #ifdef CONFIG_COMPAT
1374         p->compat_robust_list = NULL;
1375 #endif
1376         INIT_LIST_HEAD(&p->pi_state_list);
1377         p->pi_state_cache = NULL;
1378 #endif
1379         /*
1380          * sigaltstack should be cleared when sharing the same VM
1381          */
1382         if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1383                 p->sas_ss_sp = p->sas_ss_size = 0;
1384
1385         /*
1386          * Syscall tracing and stepping should be turned off in the
1387          * child regardless of CLONE_PTRACE.
1388          */
1389         user_disable_single_step(p);
1390         clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1391 #ifdef TIF_SYSCALL_EMU
1392         clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1393 #endif
1394         clear_all_latency_tracing(p);
1395
1396         /* ok, now we should be set up.. */
1397         p->pid = pid_nr(pid);
1398         if (clone_flags & CLONE_THREAD) {
1399                 p->exit_signal = -1;
1400                 p->group_leader = current->group_leader;
1401                 p->tgid = current->tgid;
1402         } else {
1403                 if (clone_flags & CLONE_PARENT)
1404                         p->exit_signal = current->group_leader->exit_signal;
1405                 else
1406                         p->exit_signal = (clone_flags & CSIGNAL);
1407                 p->group_leader = p;
1408                 p->tgid = p->pid;
1409         }
1410
1411         p->nr_dirtied = 0;
1412         p->nr_dirtied_pause = 128 >> (PAGE_SHIFT - 10);
1413         p->dirty_paused_when = 0;
1414
1415         p->pdeath_signal = 0;
1416         INIT_LIST_HEAD(&p->thread_group);
1417         p->task_works = NULL;
1418
1419         /*
1420          * Make it visible to the rest of the system, but dont wake it up yet.
1421          * Need tasklist lock for parent etc handling!
1422          */
1423         write_lock_irq(&tasklist_lock);
1424
1425         /* CLONE_PARENT re-uses the old parent */
1426         if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
1427                 p->real_parent = current->real_parent;
1428                 p->parent_exec_id = current->parent_exec_id;
1429         } else {
1430                 p->real_parent = current;
1431                 p->parent_exec_id = current->self_exec_id;
1432         }
1433
1434         spin_lock(&current->sighand->siglock);
1435
1436         /*
1437          * Process group and session signals need to be delivered to just the
1438          * parent before the fork or both the parent and the child after the
1439          * fork. Restart if a signal comes in before we add the new process to
1440          * it's process group.
1441          * A fatal signal pending means that current will exit, so the new
1442          * thread can't slip out of an OOM kill (or normal SIGKILL).
1443         */
1444         recalc_sigpending();
1445         if (signal_pending(current)) {
1446                 spin_unlock(&current->sighand->siglock);
1447                 write_unlock_irq(&tasklist_lock);
1448                 retval = -ERESTARTNOINTR;
1449                 goto bad_fork_free_pid;
1450         }
1451
1452         if (likely(p->pid)) {
1453                 ptrace_init_task(p, (clone_flags & CLONE_PTRACE) || trace);
1454
1455                 init_task_pid(p, PIDTYPE_PID, pid);
1456                 if (thread_group_leader(p)) {
1457                         init_task_pid(p, PIDTYPE_PGID, task_pgrp(current));
1458                         init_task_pid(p, PIDTYPE_SID, task_session(current));
1459
1460                         if (is_child_reaper(pid)) {
1461                                 ns_of_pid(pid)->child_reaper = p;
1462                                 p->signal->flags |= SIGNAL_UNKILLABLE;
1463                         }
1464
1465                         p->signal->leader_pid = pid;
1466                         p->signal->tty = tty_kref_get(current->signal->tty);
1467                         list_add_tail(&p->sibling, &p->real_parent->children);
1468                         list_add_tail_rcu(&p->tasks, &init_task.tasks);
1469                         attach_pid(p, PIDTYPE_PGID);
1470                         attach_pid(p, PIDTYPE_SID);
1471                         __this_cpu_inc(process_counts);
1472                 } else {
1473                         current->signal->nr_threads++;
1474                         atomic_inc(&current->signal->live);
1475                         atomic_inc(&current->signal->sigcnt);
1476                         list_add_tail_rcu(&p->thread_group,
1477                                           &p->group_leader->thread_group);
1478                         list_add_tail_rcu(&p->thread_node,
1479                                           &p->signal->thread_head);
1480                 }
1481                 attach_pid(p, PIDTYPE_PID);
1482                 nr_threads++;
1483         }
1484
1485         total_forks++;
1486         spin_unlock(&current->sighand->siglock);
1487         write_unlock_irq(&tasklist_lock);
1488         proc_fork_connector(p);
1489         cgroup_post_fork(p);
1490         if (clone_flags & CLONE_THREAD)
1491                 threadgroup_change_end(current);
1492         perf_event_fork(p);
1493
1494         trace_task_newtask(p, clone_flags);
1495         uprobe_copy_process(p, clone_flags);
1496
1497         return p;
1498
1499 bad_fork_free_pid:
1500         if (pid != &init_struct_pid)
1501                 free_pid(pid);
1502 bad_fork_cleanup_io:
1503         if (p->io_context)
1504                 exit_io_context(p);
1505 bad_fork_cleanup_namespaces:
1506         exit_task_namespaces(p);
1507 bad_fork_cleanup_mm:
1508         if (p->mm)
1509                 mmput(p->mm);
1510 bad_fork_cleanup_signal:
1511         if (!(clone_flags & CLONE_THREAD))
1512                 free_signal_struct(p->signal);
1513 bad_fork_cleanup_sighand:
1514         __cleanup_sighand(p->sighand);
1515 bad_fork_cleanup_fs:
1516         exit_fs(p); /* blocking */
1517 bad_fork_cleanup_files:
1518         exit_files(p); /* blocking */
1519 bad_fork_cleanup_semundo:
1520         exit_sem(p);
1521 bad_fork_cleanup_audit:
1522         audit_free(p);
1523 bad_fork_cleanup_policy:
1524         perf_event_free_task(p);
1525 #ifdef CONFIG_NUMA
1526         mpol_put(p->mempolicy);
1527 bad_fork_cleanup_cgroup:
1528 #endif
1529         if (clone_flags & CLONE_THREAD)
1530                 threadgroup_change_end(current);
1531         cgroup_exit(p, 0);
1532         delayacct_tsk_free(p);
1533         module_put(task_thread_info(p)->exec_domain->module);
1534 bad_fork_cleanup_count:
1535         atomic_dec(&p->cred->user->processes);
1536         exit_creds(p);
1537 bad_fork_free:
1538         free_task(p);
1539 fork_out:
1540         return ERR_PTR(retval);
1541 }
1542
1543 static inline void init_idle_pids(struct pid_link *links)
1544 {
1545         enum pid_type type;
1546
1547         for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
1548                 INIT_HLIST_NODE(&links[type].node); /* not really needed */
1549                 links[type].pid = &init_struct_pid;
1550         }
1551 }
1552
1553 struct task_struct *fork_idle(int cpu)
1554 {
1555         struct task_struct *task;
1556         task = copy_process(CLONE_VM, 0, 0, NULL, &init_struct_pid, 0);
1557         if (!IS_ERR(task)) {
1558                 init_idle_pids(task->pids);
1559                 init_idle(task, cpu);
1560         }
1561
1562         return task;
1563 }
1564
1565 /*
1566  *  Ok, this is the main fork-routine.
1567  *
1568  * It copies the process, and if successful kick-starts
1569  * it and waits for it to finish using the VM if required.
1570  */
1571 long do_fork(unsigned long clone_flags,
1572               unsigned long stack_start,
1573               unsigned long stack_size,
1574               int __user *parent_tidptr,
1575               int __user *child_tidptr)
1576 {
1577         struct task_struct *p;
1578         int trace = 0;
1579         long nr;
1580
1581         /*
1582          * Determine whether and which event to report to ptracer.  When
1583          * called from kernel_thread or CLONE_UNTRACED is explicitly
1584          * requested, no event is reported; otherwise, report if the event
1585          * for the type of forking is enabled.
1586          */
1587         if (!(clone_flags & CLONE_UNTRACED)) {
1588                 if (clone_flags & CLONE_VFORK)
1589                         trace = PTRACE_EVENT_VFORK;
1590                 else if ((clone_flags & CSIGNAL) != SIGCHLD)
1591                         trace = PTRACE_EVENT_CLONE;
1592                 else
1593                         trace = PTRACE_EVENT_FORK;
1594
1595                 if (likely(!ptrace_event_enabled(current, trace)))
1596                         trace = 0;
1597         }
1598
1599         p = copy_process(clone_flags, stack_start, stack_size,
1600                          child_tidptr, NULL, trace);
1601         /*
1602          * Do this prior waking up the new thread - the thread pointer
1603          * might get invalid after that point, if the thread exits quickly.
1604          */
1605         if (!IS_ERR(p)) {
1606                 struct completion vfork;
1607
1608                 trace_sched_process_fork(current, p);
1609
1610                 nr = task_pid_vnr(p);
1611
1612                 if (clone_flags & CLONE_PARENT_SETTID)
1613                         put_user(nr, parent_tidptr);
1614
1615                 if (clone_flags & CLONE_VFORK) {
1616                         p->vfork_done = &vfork;
1617                         init_completion(&vfork);
1618                         get_task_struct(p);
1619                 }
1620
1621                 wake_up_new_task(p);
1622
1623                 /* forking complete and child started to run, tell ptracer */
1624                 if (unlikely(trace))
1625                         ptrace_event(trace, nr);
1626
1627                 if (clone_flags & CLONE_VFORK) {
1628                         if (!wait_for_vfork_done(p, &vfork))
1629                                 ptrace_event(PTRACE_EVENT_VFORK_DONE, nr);
1630                 }
1631         } else {
1632                 nr = PTR_ERR(p);
1633         }
1634         return nr;
1635 }
1636
1637 /*
1638  * Create a kernel thread.
1639  */
1640 pid_t kernel_thread(int (*fn)(void *), void *arg, unsigned long flags)
1641 {
1642         return do_fork(flags|CLONE_VM|CLONE_UNTRACED, (unsigned long)fn,
1643                 (unsigned long)arg, NULL, NULL);
1644 }
1645
1646 #ifdef __ARCH_WANT_SYS_FORK
1647 SYSCALL_DEFINE0(fork)
1648 {
1649 #ifdef CONFIG_MMU
1650         return do_fork(SIGCHLD, 0, 0, NULL, NULL);
1651 #else
1652         /* can not support in nommu mode */
1653         return -EINVAL;
1654 #endif
1655 }
1656 #endif
1657
1658 #ifdef __ARCH_WANT_SYS_VFORK
1659 SYSCALL_DEFINE0(vfork)
1660 {
1661         return do_fork(CLONE_VFORK | CLONE_VM | SIGCHLD, 0,
1662                         0, NULL, NULL);
1663 }
1664 #endif
1665
1666 #ifdef __ARCH_WANT_SYS_CLONE
1667 #ifdef CONFIG_CLONE_BACKWARDS
1668 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
1669                  int __user *, parent_tidptr,
1670                  int, tls_val,
1671                  int __user *, child_tidptr)
1672 #elif defined(CONFIG_CLONE_BACKWARDS2)
1673 SYSCALL_DEFINE5(clone, unsigned long, newsp, unsigned long, clone_flags,
1674                  int __user *, parent_tidptr,
1675                  int __user *, child_tidptr,
1676                  int, tls_val)
1677 #elif defined(CONFIG_CLONE_BACKWARDS3)
1678 SYSCALL_DEFINE6(clone, unsigned long, clone_flags, unsigned long, newsp,
1679                 int, stack_size,
1680                 int __user *, parent_tidptr,
1681                 int __user *, child_tidptr,
1682                 int, tls_val)
1683 #else
1684 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
1685                  int __user *, parent_tidptr,
1686                  int __user *, child_tidptr,
1687                  int, tls_val)
1688 #endif
1689 {
1690         return do_fork(clone_flags, newsp, 0, parent_tidptr, child_tidptr);
1691 }
1692 #endif
1693
1694 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
1695 #define ARCH_MIN_MMSTRUCT_ALIGN 0
1696 #endif
1697
1698 static void sighand_ctor(void *data)
1699 {
1700         struct sighand_struct *sighand = data;
1701
1702         spin_lock_init(&sighand->siglock);
1703         init_waitqueue_head(&sighand->signalfd_wqh);
1704 }
1705
1706 void __init proc_caches_init(void)
1707 {
1708         sighand_cachep = kmem_cache_create("sighand_cache",
1709                         sizeof(struct sighand_struct), 0,
1710                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_DESTROY_BY_RCU|
1711                         SLAB_NOTRACK, sighand_ctor);
1712         signal_cachep = kmem_cache_create("signal_cache",
1713                         sizeof(struct signal_struct), 0,
1714                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1715         files_cachep = kmem_cache_create("files_cache",
1716                         sizeof(struct files_struct), 0,
1717                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1718         fs_cachep = kmem_cache_create("fs_cache",
1719                         sizeof(struct fs_struct), 0,
1720                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1721         /*
1722          * FIXME! The "sizeof(struct mm_struct)" currently includes the
1723          * whole struct cpumask for the OFFSTACK case. We could change
1724          * this to *only* allocate as much of it as required by the
1725          * maximum number of CPU's we can ever have.  The cpumask_allocation
1726          * is at the end of the structure, exactly for that reason.
1727          */
1728         mm_cachep = kmem_cache_create("mm_struct",
1729                         sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
1730                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1731         vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC);
1732         mmap_init();
1733         nsproxy_cache_init();
1734 }
1735
1736 /*
1737  * Check constraints on flags passed to the unshare system call.
1738  */
1739 static int check_unshare_flags(unsigned long unshare_flags)
1740 {
1741         if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
1742                                 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
1743                                 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET|
1744                                 CLONE_NEWUSER|CLONE_NEWPID))
1745                 return -EINVAL;
1746         /*
1747          * Not implemented, but pretend it works if there is nothing to
1748          * unshare. Note that unsharing CLONE_THREAD or CLONE_SIGHAND
1749          * needs to unshare vm.
1750          */
1751         if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) {
1752                 /* FIXME: get_task_mm() increments ->mm_users */
1753                 if (atomic_read(&current->mm->mm_users) > 1)
1754                         return -EINVAL;
1755         }
1756
1757         return 0;
1758 }
1759
1760 /*
1761  * Unshare the filesystem structure if it is being shared
1762  */
1763 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
1764 {
1765         struct fs_struct *fs = current->fs;
1766
1767         if (!(unshare_flags & CLONE_FS) || !fs)
1768                 return 0;
1769
1770         /* don't need lock here; in the worst case we'll do useless copy */
1771         if (fs->users == 1)
1772                 return 0;
1773
1774         *new_fsp = copy_fs_struct(fs);
1775         if (!*new_fsp)
1776                 return -ENOMEM;
1777
1778         return 0;
1779 }
1780
1781 /*
1782  * Unshare file descriptor table if it is being shared
1783  */
1784 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
1785 {
1786         struct files_struct *fd = current->files;
1787         int error = 0;
1788
1789         if ((unshare_flags & CLONE_FILES) &&
1790             (fd && atomic_read(&fd->count) > 1)) {
1791                 *new_fdp = dup_fd(fd, &error);
1792                 if (!*new_fdp)
1793                         return error;
1794         }
1795
1796         return 0;
1797 }
1798
1799 /*
1800  * unshare allows a process to 'unshare' part of the process
1801  * context which was originally shared using clone.  copy_*
1802  * functions used by do_fork() cannot be used here directly
1803  * because they modify an inactive task_struct that is being
1804  * constructed. Here we are modifying the current, active,
1805  * task_struct.
1806  */
1807 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
1808 {
1809         struct fs_struct *fs, *new_fs = NULL;
1810         struct files_struct *fd, *new_fd = NULL;
1811         struct cred *new_cred = NULL;
1812         struct nsproxy *new_nsproxy = NULL;
1813         int do_sysvsem = 0;
1814         int err;
1815
1816         /*
1817          * If unsharing a user namespace must also unshare the thread.
1818          */
1819         if (unshare_flags & CLONE_NEWUSER)
1820                 unshare_flags |= CLONE_THREAD | CLONE_FS;
1821         /*
1822          * If unsharing a thread from a thread group, must also unshare vm.
1823          */
1824         if (unshare_flags & CLONE_THREAD)
1825                 unshare_flags |= CLONE_VM;
1826         /*
1827          * If unsharing vm, must also unshare signal handlers.
1828          */
1829         if (unshare_flags & CLONE_VM)
1830                 unshare_flags |= CLONE_SIGHAND;
1831         /*
1832          * If unsharing namespace, must also unshare filesystem information.
1833          */
1834         if (unshare_flags & CLONE_NEWNS)
1835                 unshare_flags |= CLONE_FS;
1836
1837         err = check_unshare_flags(unshare_flags);
1838         if (err)
1839                 goto bad_unshare_out;
1840         /*
1841          * CLONE_NEWIPC must also detach from the undolist: after switching
1842          * to a new ipc namespace, the semaphore arrays from the old
1843          * namespace are unreachable.
1844          */
1845         if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
1846                 do_sysvsem = 1;
1847         err = unshare_fs(unshare_flags, &new_fs);
1848         if (err)
1849                 goto bad_unshare_out;
1850         err = unshare_fd(unshare_flags, &new_fd);
1851         if (err)
1852                 goto bad_unshare_cleanup_fs;
1853         err = unshare_userns(unshare_flags, &new_cred);
1854         if (err)
1855                 goto bad_unshare_cleanup_fd;
1856         err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
1857                                          new_cred, new_fs);
1858         if (err)
1859                 goto bad_unshare_cleanup_cred;
1860
1861         if (new_fs || new_fd || do_sysvsem || new_cred || new_nsproxy) {
1862                 if (do_sysvsem) {
1863                         /*
1864                          * CLONE_SYSVSEM is equivalent to sys_exit().
1865                          */
1866                         exit_sem(current);
1867                 }
1868
1869                 if (new_nsproxy)
1870                         switch_task_namespaces(current, new_nsproxy);
1871
1872                 task_lock(current);
1873
1874                 if (new_fs) {
1875                         fs = current->fs;
1876                         spin_lock(&fs->lock);
1877                         current->fs = new_fs;
1878                         if (--fs->users)
1879                                 new_fs = NULL;
1880                         else
1881                                 new_fs = fs;
1882                         spin_unlock(&fs->lock);
1883                 }
1884
1885                 if (new_fd) {
1886                         fd = current->files;
1887                         current->files = new_fd;
1888                         new_fd = fd;
1889                 }
1890
1891                 task_unlock(current);
1892
1893                 if (new_cred) {
1894                         /* Install the new user namespace */
1895                         commit_creds(new_cred);
1896                         new_cred = NULL;
1897                 }
1898         }
1899
1900 bad_unshare_cleanup_cred:
1901         if (new_cred)
1902                 put_cred(new_cred);
1903 bad_unshare_cleanup_fd:
1904         if (new_fd)
1905                 put_files_struct(new_fd);
1906
1907 bad_unshare_cleanup_fs:
1908         if (new_fs)
1909                 free_fs_struct(new_fs);
1910
1911 bad_unshare_out:
1912         return err;
1913 }
1914
1915 /*
1916  *      Helper to unshare the files of the current task.
1917  *      We don't want to expose copy_files internals to
1918  *      the exec layer of the kernel.
1919  */
1920
1921 int unshare_files(struct files_struct **displaced)
1922 {
1923         struct task_struct *task = current;
1924         struct files_struct *copy = NULL;
1925         int error;
1926
1927         error = unshare_fd(CLONE_FILES, &copy);
1928         if (error || !copy) {
1929                 *displaced = NULL;
1930                 return error;
1931         }
1932         *displaced = task->files;
1933         task_lock(task);
1934         task->files = copy;
1935         task_unlock(task);
1936         return 0;
1937 }